Keep container logs writing after an attached client dies

[rogerneal]

Summary

• Isolate per-handle write failures in MultiWriter so a dead attached client (EPIPE) does not stop writes to the on-disk container log.
• Drop handles that fail so later writes do not keep hitting the broken pipe.
• Move MultiWriter into its own file and add unit tests that reproduce the EPIPE fan-out failure.

Fixes #2009

Test plan

Tested

• swift test --filter MultiWriterTests
  • writeContinuesToLogAfterClientEPIPE: client pipe closed (EPIPE), log file still receives subsequent writes
  • writeThrowsWhenEveryHandleFails: throws when no handles remain writable
• Full CLI reproduction with a build of this branch installed under /tmp:
  1. container run --name freeze alpine:latest sh -c 'while true; do echo "tick $(date -u +%H:%M:%S)"; sleep 2; done'
  2. Wait until running with at least 3 ticks (before_count=3, last tick 00:07:31)
  3. kill -9 the attached container run PID
  4. Wait 14s; container stayed running
  5. container logs freeze showed 7 additional ticks through tick 00:07:45 (after_count=10)
• Tests ignore SIGPIPE the same way RuntimeLinuxHelper does, so broken-pipe writes surface as EPIPE